CellFlux: Simulating Cellular Morphology Changes via Flow Matching

Yuhui Zhang, Yuchang Su, Chenyu Wang, Tianhong Li, Zoe Wefers, Jeffrey J Nirschl, James Burgess, Daisy Ding, Alejandro Lozano, Emma Lundberg, Serena Yeung-Levy

TL;DR: CellFlux is an image-generative model that simulates cellular morphology changes from chemical and genetic perturbations using flow matching.

Abstract: Building a virtual cell capable of accurately simulating cellular behaviors in silico has long been a dream in computational biology. We introduce CellFlux, an image-generative model that simulates cellular morphology changes induced by chemical and genetic perturbations using flow matching. Unlike prior methods, CellFlux models distribution-wise transformations from unperturbed to perturbed cell states, effectively distinguishing actual perturbation effects from experimental artifacts such as batch effects—a major challenge in biological data. Evaluated on chemical (BBBC021), genetic (RxRx1), and combined perturbation (JUMP) datasets, CellFlux generates biologically meaningful cell images that faithfully capture perturbation-specific morphological changes, achieving a 35% improvement in FID scores and a 12% increase in mode-of-action prediction accuracy over existing methods. Additionally, CellFlux enables continuous interpolation between cellular states, providing a potential tool for studying perturbation dynamics. These capabilities mark a significant step toward realizing virtual cell modeling for biomedical research. Project page: https://yuhui-zh15.github.io/CellFlux/.

Lay Summary: Scientists have long dreamed of creating a “virtual cell”—a computer model that can accurately simulate how real cells respond to drug or genetic perturbations. Our work introduces CellFlux, a new image-generative model that simulates how cell shape changes under different experimental conditions. Unlike previous methods that directly generate perturbed cell images, CellFlux reformulates the problem as a transformation between two distributions: unperturbed and perturbed cells. It learns this transformation using a principled technique called flow matching. This formulation and solution address a fundamental challenge in biology: batch effects, which—similar to distribution shifts in machine learning—often confound true biological signals with experimental noise. We tested CellFlux on three chemical and genetic perturbation datasets and found that it produces more accurate and biologically meaningful cell images than previous methods. It improves image quality by 35% and biological metrics by 12%. Even more excitingly, CellFlux can simulate smooth and continuous transitions in cell shape between different states—like watching a time-lapse of a cell responding to treatment—providing a valuable tool for studying perturbation dynamics. Overall, CellFlux marks a significant step toward realizing virtual cell modeling for biomedical research.

Application-Driven Machine Learning: This submission is on Application-Driven Machine Learning.

Link To Code: https://github.com/yuhui-zh15/CellFlux

Primary Area: Applications->Health / Medicine

Keywords: flow matching, cell image, drug discovery, generative models

Submission Number: 393